RegulationModels 🇺🇸 24.07.2026 01:02

White House adviser accuses Chinese Moonshot of stealing Anthropic model, experts skeptical

Moonshot AIMoonshot AI AnthropicAnthropic
White House science advisor Michael Kratsios accused Chinese AI company Moonshot of copying Anthropic's Fable LLM through distillation and using banned chips. Experts question whether distillation alone could produce Kimi K3's capabilities in such a short time, noting the technical expertise of Chinese teams and common industry practices.
White House science advisor Michael Kratsios claimed that Moonshot, the Chinese company behind the open-weight LLM Kimi K3, built its model by distilling Anthropic's Fable LLM while using Nvidia Grace Blackwell 300 chips not cleared for export to China. He wrote on social media that such covert industrial distillation is unacceptable, amid discussions about banning Chinese open-weight models. Treasury Secretary Scott Bessent echoed that watermarks of U.S. models are found on Chinese models. However, experts are skeptical: Braden Hancock of Laude Institute said there wasn't enough time since Fable's July 1 release to distill, train, and release Kimi K3 in two weeks. Nathan Lambert of Allen Institute for AI noted that distillation is becoming less impactful as models advance, and achieving Fable-like capabilities would require reinforcement learning with significant infrastructure, making API-based distillation prohibitively expensive and slow. Anthropic previously accused Moonshot, DeepSeek, and MiniMax of systematic distillation, citing millions of unusual queries. Yet distillation is common; Elon Musk testified that his company distilled OpenAI models for Grok. Hancock emphasized the technical expertise of Chinese teams, noting Moonshot's founder was a CMU PhD student. Kratsios also alleged Moonshot accessed advanced chips via a black market, as Supermicro's founder was indicted for smuggling. Proposed know-your-customer rules for data centers have stalled under the Trump administration.
Abbreviations
LLM = Large Language Model — большая языковая модель
SFT = Supervised Fine-Tuning — контролируемая донастройка
API = Application Programming Interface — интерфейс программирования приложений
CMU = Carnegie Mellon University — Университет Карнеги — Меллон
Source: TechCrunch AI — original
Our earlier posts on this topic ↓
Fresh news